Age differences in the principal temporo-spatial components of EEG activity during a proactive interference task

.


Introduction
Working memory (WM) is the ability to maintain and manipulate information in the mind over a short period of time (Baddeley, 2010).It has been found to suffer a marked decline in healthy aging (Park et al., 2002), which the inhibitory deficit hypothesis reasons may be due to a decreased ability to inhibit disruption from goal-irrelevant information (Hasher & Zacks, 1988).One situation requiring such inhibitory control is the build-up of proactive interference (PI).PI is the disruptive effect of no longer relevant information on current WM processing (Jonides & Nee, 2006), and has been found to be predictive of WM capacity (May et al., 1999).Resistance to PI is an inhibition-related function (Friedman & Miyake, 2004); hence, the study of PI effects may help understand how inhibitory control of WM contents is associated with age-related WM decline.
PI has been previously researched using the Recent Probes task (e.g., Jonides et al., 2000).In this task, participants must memorise an array of stimuli (Target Set) to decide whether an individual stimulus presented a few seconds later (Probe) was part of that array.The Probes that produce the highest levels of PI do not match any stimulus in the Target Set of the current trial but do match a stimulus in the Target Set of the previous trial.These Probes, called recent negative (RN) Probes, are highly familiar and cause a tendency to produce a positive response because of their recent relevance for the task (i.e., in the previous trial).However, in the current trial, RN Probes are irrelevant, meaning that to correctly respond to them participants must recollect the context in which they were previously encountered (i.e., in the previous trial) and provide a negative response.PI has been suggested to arise from the conflict between RN Probes' high familiarity and the recollection of the context in which they were previously encountered (Oberauer, 2005).
PI effects in the Recent Probes task are generally observed as longer reaction times (RTs) on RN compared to non-recent negative (non-RN) Probes (D'Esposito et al., 1999).The later stimuli are not presented in the current or previous trial, producing low levels of PI.Other studies have also reported PI effects in response accuracy as well as in RTs (Badre & Wagner, 2005;Jonides et al., 1998;Llorens et al., 2020;Loosli et al., 2014;Loosli et al., 2016;Mecklinger et al., 2003;Nee et al., 2007;Nelson et al., 2003;Zhang et al., 2010).
PI effects in the Recent Probes task have also been observed at the brain activity level using electroencephalographic data (EEG).Zhang et al. (2010) used the event-related potential (ERP) technique to analyse EEG data recorded during the Recent Probes task in young adults (YA).They found a Late Positive Component (LPC) at centro-parietal electrodes to have greater amplitude in non-RN compared to RN trials between 400 and 600 ms after Probe onset.These authors argue that this PI effect reflects the competition between familiarity and recollection of contextual information during PI.This result was partially replicated by Calvo and Bialystok (2021), who observed an LPC to be modulated by PI during a slightly later period between 650 and 850 ms after Probe onset in a Recent Probes task with verbal stimuli.LPC has also been observed in other WM tasks requiring participants to make perceptual decisions on stimuli that may produce interference, as well as by incongruent stimuli in the Stroop paradigm (Vo et al., 2021).However, the LPC in Stroop paradigms has been reported at later time windows between 460 and 1200 ms post stimulus (Vo et al., 2021).
In addition to LPC, Zhang et al. (2010) observed a frontal negativity peaking between 250-350 ms after Probe presentation (control-related anterior N2), which was sensitive to response type (positive or negative) but not to PI condition.In contrast, Llorens et al. (2020) failed to observe PI effects in LPC but found a control-related anterior N2 to be modulated by recency, showing larger amplitude for RN than non-RN trials, in accordance with the suggested sensitivity of this component to the extent of discrepancy and response conflict induced by PI (see Du et al., 2008;Folstein & Van Petten, 2008).Furthermore, Calvo and Bialystok (2021) also observed a control-related anterior N2 to be modulated by PI in their verbal Recent Probes task.In addition, similar PI modulations to those reported for this control-related anterior N2 (Llorens et al., 2020) were observed for a later medial frontal negativity (MFN) peaking around 420 ms (Tays et al., 2008), which has also been observed for PI induced by 2 and 4-back lures in a 3-back task (Perfetti et al., 2014).Thus, MFN is thought to reflect the extent to which cognitive control is recruited to overcome interference or conflict.
Regarding differences in PI effects between YA and healthy older adults (OA), a pair of complementary studies by Jonides and colleagues (Jonides et al., 1998;Jonides et al., 2000) found OA to show larger PI effects than YA when considering both accuracy and RT.Tays et al. (2008), found PI effects to be present in RT regardless of age, but only in OA's response accuracy.These authors have also found age-related differences in PI effects on EEG signals.They observed larger MFN amplitude for RN than non-RN trials in YA, but no such difference in OA, who showed a slight positive deflection in this same time window (i.e.400 − 500 ms post-stimulus) and region.These authors suggested that YA modulate their brain response according to the cognitive demands of the different PI conditions, while OA are incapable of such selective recruitment.This pattern of results is consistent with cognitive aging theories suggesting that OA recruit neural resources less efficiently when cognitive control is required (Grady, 2008;Reuter-Lorenz & Park, 2014).Yi and Friedman (2014) also found age-related differences in PI effects, although using a modified Sternberg WM task.In this task, between the Target Set and Probe presentation a visual cue is presented to indicate which Target Set's stimuli are irrelevant to the task.High PI is produced when the Probe matches one of the Target Set's stimuli signaled as irrelevant.The authors found the difference in RTs between high and low PI trials (i.e., PI cost) to be greater in OA than in YA.Furthermore, YA elicited a positive component at fronto-central electrodes between 300 and 500 ms, while OA elicited a similar component, albeit between 600 and 800 ms.Similarly to the LPC found by Zhang et al. (2010) in YA, this positive component had higher amplitude for low compared to high PI trials in both age groups.
Age-related latency differences suggest that OA may have reduced processing speed, a factor known to hinder OA's performance in WM tasks (Salthouse, 1994).However, as Yi and Friedman (2014) did not find neural effects to be correlated with behaviour, it was not possible to determine to what extent the reduced processing speed at the neural level contributed to the PI effects at the behavioural level.Alternatively, the observed age-related differences in ERP components may signal variations in the structure of brain activity and mental processing, which also explains Tays and colleagues' (2008) findings.Thus, it can be argued that it is unclear whether the age-related differences in ERP waveforms observed in Recent Probes tasks represent increased latency in OA compared to YA, indicative of a reduced processing speed, or whether distinct ERP components were evoked in each group.
Further, the different patterns of behavioural and physiological results reported in previous studies may be explained by methodological differences.For instance, whereas Yi and Friedman (2014) used a balanced proportion of PI-evoking trials, Tays et al. (2008) used a relatively high percentage of PI-inducing trials, increasing the likelihood of participants' awareness of the PI manipulation.
Consequently, the present study aimed to further knowledge on this topic by having YA and OA complete a Recent Probes task with a balanced distribution of conditions across trials and by applying an analytical technique that allowed us to identify the constituent components of the ERPs and separate potentially overlapping ERP components.Principal component analysis (PCA) was, thus, used to achieve these objectives.PCA is a factor analytic technique that extracts linear combinations of latent variables using eigenvalue decomposition (Dien & Frishkoff, 2005).It is a data-driven method to extract variance in the EEG signal in a precise, unbiased, and parsimonious way by offering solutions for separating variance in ERP components that may be otherwise confounded by the naked eye (Dien, 2010).Therefore, applying temporo-spatial PCA (tsPCA) to EEG data recorded during the resolution of PI would help untangle potentially overlapping ERP components and enhance our knowledge of how the neural correlates of cognitive control change in healthy aging.We expected to find PI effects in YA's and OA's accuracy and RTs.Still, PI costs would be larger in OA compared to YA, supporting the inhibitory deficit hypothesis of healthy aging (Hasher & Zacks, 1988).Based on the reviewed results, we expected to observe PI effects in EEG data in a negative deflection between 250 and 550 ms and a positive deflection between 400 and 1200 ms after Probe stimulus onset in fronto-central and centro-parietal electrodes, with greater positivity in conditions of low (i.e., non-RN) compared to high (i.e., RN) PI.However, these EEG data effects should be altered in OA.

Participants
Participants were recruited via a course credit system for an undergraduate Psychology course or an invitation to OA enrolled in senior universities and leisure centres, which were offered a report on a complete neuropsychological assessment as compensation.Participants gave full written informed consent and were treated following the principles stated in the Declaration of Helsinki.The study's protocols were approved by the Ethics Committee for Life and Health Sciences at the University of Minho (Ref. SECVS 108/2016).
Participants' data was excluded from analyses if they had: a) history of or an active pharmacological or psychotherapeutic treatment for a psychiatric or neurological disorder (N = 1); or b) a score above 13 points in the Beck Depression Inventory (BDI; Beck et al., 1996) suggesting moderate or severe depressive symptomatology (N = 2).Further, all OA scored within two standard deviations of the normative score for their age and education level in the Montreal Cognitive Assessment (MoCA; Nasreddine et al., 2005).One participant was excluded due to incomplete data.Additionally, five participants were excluded due to outlying scores in the behavioural performance data (see below for details) and another two due to numerous EEG artefacts.In total, eight YA H.T. A. Moore et al. and three OA were excluded from the analyses.
Using the lower effect size for a (marginally) significant interaction of PI condition (RN vs Non-RN) Group (young and old adults) in a repeated measure ANOVA on behavioural data reported by Loosli et al. (2016) (i.e., η p 2 =.10), G*Power 3.1 (Faul et al., 2009) was used to calculate the required sample size to detect an effect of that size in an ANOVA with the same design,.80power, an alpha level of 0.05 and assuming low correlation among the repeated measures (r = 0.5).Results indicated a required total sample size of 20 participants.However, twenty-six YA between 18 and 34 years (16 females) and sixteen OA between 53 and 68 years (9 females) were included in the study.Participant data regarding years of education and handedness are also presented in Table 1.

Procedure
The stimuli presented in the Recent Probes task were 96 emotionally neutral images: 48 faces (24 female) from the Karolinska Directed Emotional Faces (KDEF; Lundqvist et al., 2004) and 48 scenes from the International Affective Picture System (IAPS; Lang, Bradley & Cuthbert, 2008).The values regarding the emotionality of the IAPS and KDEF images used can be seen in Supplementary Tables 1 and 2, respectively.Stimuli were presented on a computer screen (21.5", 60 Hz refresh rate) in colour on a black background.
The Recent Probes task was implemented using PsychoPy2 v1.84 (Peirce, 2007).Participants were presented with a Target Set of six images (three faces and three scenes) around a white fixation cross located in the centre of the screen (two faces and one scene above and two scenes and one face below the fixation cross, or vice versa) for 2000 ms.Then, only the white fixation cross remained visible during a delay period jittered between 2750-3250 ms across trials (average = 3000 ms).After the delay period, the Probe image (either a single face or scene) was presented in the centre of the screen for 2000 ms.Then, the white fixation cross was presented again in the centre of the screen for an average of 3000 ms (jittered between 2500-3500 ms across trials) until the beginning of the following trial (see Fig. 1).
Trials were divided into four conditions according to the required response and Probe recency: recent positives (RP), non-recent positives (non-RP), RNs, and non-RNs.Regarding response type, both positive (P) trial conditions required a "yes" response since the Probe matched a stimulus in the Target Set of the current trial.In contrast, trials in the negative conditions (N) required a "no" response.Regarding recency, both recent (R) trial conditions had a Probe that matched a stimulus in the Target Set of the previous trial.In contrast, the Probes in trials of the non-recent (non-R) conditions did not match any stimuli in the Target Set of, at least, the six previous trials.Each condition comprised 25 % of all trials in each task block (see below).Conditions were presented randomly across the block but for the first trial.The Recent Probes paradigm is limited in this regard since the first trial of each block can only be randomly selected among the non-RP and non-RN conditions since the Probe stimulus cannot be presented before the first trials' target set.Given that the ERP waveforms between the conditions of interest (i.e.non-RN and RN) were similar and the relatively low number of trials (see below), the first trial of each block were not discarded from analyses.Note that all Target Sets contained three stimuli (two faces and one scene, or vice versa) that were also presented in the previous trial and three stimuli that were not shown in the last six trials to avoid confounding familiarity with PI.Also, the Probe stimulus in a trial was not used as Probe stimulus for at least another thirty-two trials.
The task was completed in two blocks of 96 trials.In one block, the Probe was always a face, and participants had to selectively memorise the three face stimuli in the Target Set.In the other block, the Probe was always a scene, and participants had to selectively memorise the three scene stimuli in the Target Set.Face and scene stimuli were presented together in the Target Set to increase the ecological validity of the task by having irrelevant stimuli intermixed with relevant information.The order of block completion was randomised across participants.Participants completed the task in a soundproof cabin and were instructed to respond as accurately and quickly as possible.They responded with their dominant hand by pressing a button for "yes" and a button for "no".The response keys (i.e., 4 and 6 in a numeric pad) were counterbalanced across participants.

Data analysis
We analysed accuracy and RT across both blocks only for trials in the RN and non-RN conditions.Accuracy was measured as the percentage of trials with correct "no" responses, and RT as the time elapsed from Probe stimulus onset to "no" button press for correct responses.Further, proportional PI costs for RT and accuracy rates were calculated separately using the following formula: (RNnon-RN)/non-RN, where RN stands for RT or accuracy rate for the RN condition and non-RN for the corresponding parameter in the non-RN condition.Additionally, sensitivity (d') and response bias (c) parameters from signal detection theory were calculated according to the formulas presented by Stanislaw and Todorov (1999).Participants were excluded from analysis if their behavioural data on any variable of interest had a z-score larger than ± 3.
Robust Analyses of Variance (ANOVAs) with the within-subjects factor PI Condition (non-RN vs RN) and with the between-subjects factor Age Group (YA vs OA) were run on accuracy, RTs, d' and c.Following the statistical procedures described by Keselman et al., (2003Keselman et al., ( , 2008)), to avoid the biasing effects of non-normality and (co)variance heterogeneity, these robust inferential tests used a Welch-James statistic computed using trimmed means (0.05 % of each data distribution extreme) and winsorized variances/covariances.The critical values for the Welch-James statistic were calculated using 4999 bootstrap samples in the EP Toolkit v2.992 (Dien, 2010).Cohen's d-type effect sizes and their 95 % confidence intervals for the Welch-James statistic of the robust ANOVAs were calculated using the welchADF package (Villacorta, 2017) in R v1.1.419(R Core Team, 2021) following the equation proposed by Keselman and colleagues (2008).The means reported for all robust ANOVAs are trimmed means, which are reported with their corresponding non-winsorized standard deviations (SDs).Finally, independent samples t-tests were used to compare proportional PI costs for RT and accuracy between groups in JASP Version 0.16 (JASP Team, 2023).The means reported for these t-tests are non-trimmed means, which are reported with their corresponding non-winsorized SDs.Alpha level was set at 0.05 for all statistical tests.

EEG recording and analyses
EEG signal was recorded using an ActiveTwo Biosemi system (Biosemi, Amsterdam, the Netherlands) with 64 active electrodes inserted in  (Oldfield, 1971).
an elastic cap according to the international 10-10 system.A Common Mode Sense (CMS) and Driven Right Leg (DRL) montage of two electrodes around the vertex was used as a reference.Electrode offset was kept below 30 mV.EEG data was filtered online between 0.01-100 Hz and digitised at a sampling rate of 512 Hz.Only correctly answered negative trials (i.e., non-RN and RN trials) entered EEG analyses, which were run in ERPLab v7.0 (Lopez-Calderon & Luck, 2014) and EEGLab v14.0.0 (Delorme & Makeig, 2004).Data was filtered offline following the recommendations on ERPLab plugin documentation for band-pass filtering.Thus, a two-pass Butterworth filter between 0.1 and 30 Hz with 12 dB roll-off was applied to continuous data.Bad channels were eliminated, and transient, non-stereotyped, large amplitude artefacts likely to be of muscular or technical (e.g.sensor displacement) origin were corrected using Artefact Subspace Reconstruction (see Mullen et al., 2015) as implemented by the Clean Rawdata plugin (v.0.32) for EEGlab.The EEG signal was re-referenced to the average and segmented from 200 ms before to 1500 ms after Probe stimulus onset.The pre-stimulus interval mean voltage was subtracted from each channel's entire waveform for baseline correction in each epoch.Independent component (IC) analysis was performed to remove additional ocular and muscular artefacts by eliminating the ICs with a high likelihood (i.e., > 70 %) of corresponding to ocular, cardiac, or muscular activity as suggested by the classification made by the ICLabel v0.3 plugin for EEGlab (Pion-Tonachini et al., 2019).Eliminated channels were interpolated using superfast spherical interpolation.Automated artefact detection routines rejected epochs with data points exceeding ± 150 μV or two consecutive data points that differed by more than 50 μV.The total number of epochs removed due to artefacts did not differ between age groups, t(40) = − 1.04, p = .30(YA: M =.11, SD =.49; OA: non-trimmed Mean =.30, SD =.70).Epochs were averaged at each electrode according to PI condition (non-RN and RN) across blocks, resulting in 2 bins of averaged ERP waveforms for each participant.There were no between group differences in the number of epochs that entered analysis in either condition (see Table 2).
We used the EP Toolkit v2.992 (Dien, 2010) to apply a tsPCA to the ERPs.PCA was carried out between − 200 and 960 ms relative to Probe onset.The latter corresponds with the mean RT of YA across all negative trials (see Results section).PCA is a dimensionality reduction technique that forms linear combinations of the original measurements that capture most of the relevant variance in factors that reflect latent ERP patterns (for details, see Cohen, 2014;Dien & Frishkoff, 2005;Picton et al., 2000).Hence, to determine whether age groups showed differences in the time courses of the latent ERP patterns elicited by Probe stimuli, we carried out tPCA separately on each age group (see Supplementary Material for a tsPCA in OA between − 200 and 1112 ms relative to Probe onset).
ERP waveforms at each electrode for each participant, PI condition, and task block were entered into a tPCA done on the covariance matrix with no factor rotation.A parallel test (Horn, 1965) was applied to decide the number of factors to retain (see Dien, 2012), which were then submitted to a Promax factor rotation with Kaiser loading weighting.In YA, this procedure resulted in twenty-two temporal factors that explained 96.21 % of the total variance (see Fig. S1A).tPCA in OA resulted in twenty temporal factors that explained 94.45 % of the total variance (see Fig. S1B).
As the resulting temporal components in both age groups differed considerably, we proceeded with sPCA on each group separately.An sPCA was run on the factor scores for each retained temporal factor separately, resulting in one eigen-decomposition (i.e., vector of eigenvalues) per temporal factor.Each element of these vectors represents an unrotated spatial factor.Results of a parallel test to determine how many components to retain suggested four spatial factors to be retained at each temporal factor.The suggested number of spatial factors for each temporal factor were submitted to an Infomax factor rotation, resulting in eighty-eight temporo-spatial factors (TSFs) in YA and eighty in OA.The total variance explained by each TSF was calculated, and only those TSFs explaining at least 0.5 % of the total variance were retained for further analyses (see Dien & Frishkoff, 2005;Dien, 2012).As a result, twenty-two TSFs explaining 61.48 % of total data variance for YA (see Table S3 and Fig. S2) and twenty-four TSFs explaining 62.51 % of total data variance for OA (see Table S3 and Fig. S3) were selected for further analyses.
As recommended by Dien (2012), a two-step approach for statistical analyses was followed.First, a targeted analysis was run on those TSFs that showed similar scalp topography and occurred in the same temporal window as the ERP components observed in previous studies on PI effects: Tays et al. (2008) observed PI and age modulations on an MFN between 400 to 500 ms post stimulus in a cluster of frontal electrodes centred around and including AFz.Further, given that previous studies have found the PI induced by Probe stimuli to modulate control-related N2-like negative ERP components at earlier time windows, such as 200 to 250 ms (Du et al., 2008) and between 250 to 350 ms (Zhang et al., 2010), we opted to focus our analysis for medial frontal negativities on the time window from 250 ms to 550 ms after Probe onset.Zhang et al. (2010) observed an LPC between 400 and 600 ms post stimulus at central and parietal electrodes to be modulated by PI.However, other studies have observed PI tasks to elicit LPCs at fronto-central electrodes between 600 and 800 ms in OA (Yi & Friedman, 2014) or between 650 and 850 ms in YA (Calvo & Bialystok, 2021), and at much later time windows in WM tasks that provoke Stroop-like interference (i.e.460 to 1200 ms, Vo et al., 2021).Therefore, we selected TSFs for the targeted analysis that showed late fronto-central or parietal positivities between 400 and 960 ms post stimulus.
Hence, TSFs 1.1, 6.1, and 8.1 for YA and TSFs 3.1 and 10.1 for OA were selected based on the similarity of their spatial and temporal characteristics with MFN and MFN-like components modulated by PI in previous studies.On the other hand, TSFs 2.2, and 3.2 in YA and TSFs 1.2 and 8.1 for OA were chosen based on their similarity to the LPCs showing PI or similar interference modulations in previous studies.A robust ANOVA with the within-subjects factor PI condition (non-RN and RN) was conducted to analyse PI effects in each selected TSF in the corresponding age group.The factor scores of each TSF were converted to microvolts before being submitted to the robust ANOVA following the equation by Dien and Frishkoff (2005).The conversion is also described in the Supplementary Materials.Converting the factor scores to microvolts facilitates comparisons between the TSFs and the original grand average ERP waveform (Dien, 2012).The robust ANOVA followed the same procedure as those run on behavioural data.
The second step included an exploratory analysis on all remaining TSFs for YA and OA.The reconstructed voltages of these factors were submitted to a robust ANOVA of the same characteristics as the one run for the selected TSFs for the target analysis but with the alpha level adjusted following the Benjamini-Hochberg procedure to control for multiple testing (Benjamini & Hochberg, 1995).
Note that Block type was not in the scope of the present study.However, see Supplementary Materials for an exploration of block effects and their interaction on the TSFs showing significant differences between PI conditions.There was no interaction showing differences in the PI effects between blocks.
Finally, an exploratory correlational analysis was performed for those TSFs showing significant differences between PI conditions in a given age group.Thus, Pearson correlations between the TSF's reconstructed voltage difference between PI conditions (RNnon-RN) and PI costs for RT and accuracy were run in the corresponding age-group.All correlations were carried out in JASP Version 0.16 with the alpha level set at 0.05.

Source localisation analysis
Additionally, for those TSFs showing significant differences between PI conditions, standardised low-resolution brain electromagnetic tomography (sLORETA; Pascual-Marqui, 2002) was used to localise for each group separately, a plausible source of the TSF scores collapsed across conditions in LORETA-KEY v20171101 software.sLORETA offers a solution to the inverse problem by estimating the current source density (CSD) distribution for the whole brain based on the EEG activity measured at the scalp.sLORETA conducts source localisation by dividing the estimated CSD at each voxel by the square root of its estimated variance, meaning that the estimated CSD is provided in z-scores.The weights used by sLORETA rely on the assumption that neighbouring populations of neurons are highly correlated to create images representing the smoothest of all possible 3-dimensional CSD estimations.Such images have low spatial resolution and offer a plausible generator for the scalp distribution of reconstructed voltages for a given TSF.
We first converted TSF scores to voltages (from − 200 to 960 ms relative to Probe onset) and submitted them to sLORETA, using a transformation matrix created from the electrode coordinates, choosing no regularisation method and sLORETA algorithm.Note that the spatial distribution of TSF scores converted to voltages does not vary over time.Source localisation was done using a realistic head model based on the Montreal Neurological Institute (MNI152) template (Mazziotta et al., 2001).The anatomical space in which the solution was estimated comprised cortical grey matter and intracerebral volume, and was partitioned in 6239 voxels of 5 × 5 x 5 mm.Results are reported in MNI coordinates.Source localisations of the TSFs selected a priori that did not show significant differences between PI conditions can be viewed in Table S4.

Accuracy
There was a significant main effect of PI Condition on accuracy (see Fig. 2), T WJt /c(1.0,27.8) = 91.75,p < 0.001, effect size = 2.51, 95 % CI [1.87 3.95], with higher accuracy for non-RN (M = 96.07,SD = 4.73) than RN trials (M = 88.13,SD = 6.83).No main effects of Age Group or interaction effects between PI and Age Group were observed in accuracy.Age groups did not significantly differ in proportional PI accuracy costs, t( 40

Targeted and exploratory analyses in young adults
In YA, one TSF that was selected for the targeted analyses for having temporal and spatial characteristics similar to the MFN -TSF1.1showedsignificant effects of PI Condition (see Table 3 and Fig. 3).No TSFs selected based on their similarities with the LPC showed significant effects of PI condition (see Fig. 4).
Exploratory analyses of TSFs in OA did not show any effects of PI Condition on any of the TSFs.

Exploratory brain-behaviour correlational analyses
The exploratory correlation analyses between the TSFs showing significant effects and the behavioural variables showed there were no significant correlations (see Table S5 for a complete description of correlations).

Discussion
The current study aimed to contribute to understanding the effects of PI on behaviour and brain activity in healthy YA and OA.We also sought to overcome the limitations of previous work by presenting participants with a Recent Probes task with a balanced distribution of conditions across trials and applying PCA to the ERP data recorded during the task to better characterise YA's and OA's brain activity associated with PI resolution.
We expected to find PI effects in YA's and OA's accuracy and RTs, but that PI costs would be larger in OA than YA, supporting the inhibitory deficit hypothesis of healthy aging (Hasher & Zacks, 1988).This hypothesis predicts that we experience more interference from irrelevant information as we age.As expected, YA and OA responded less accurately and slower to RN than non-RN trials.Also, despite the absence of age differences in response precision, OA were generally slower to respond than YA independently of PI condition.However, in contrast to our inhibitory deficit hypothesis-derived prediction, YA showed greater RT PI costs than OA.This may be related to the difficulty or load that non-RN trials already imposed on OA, who were generally slower and showed a greater PI cost in accuracy, albeit not significantly.This latter factor, together with the lower discriminability between positive and negative Probes shown by OA (see below) leaves open the possibility that the lower PI cost in OA's RT compared with YA is related to a different approach to task resolution.In this sense, they may have prioritised accuracy over RT when completing the task to a greater extent than YA.Nonetheless, further exploration of this effect in future studies is needed to assess such potential explanations, for instance, with Recent Probes tasks with and without response time constraints or manipulating the priority of accuracy and speed in the instructions provided to participants.
As mentioned above, differences between YA and OA were also observed in the ability to discriminate between positive and negative trials (sensitivity: d'), such that young adults showed greater sensitivity than OA.It was also easier for all participants, regardless of age group, to discriminate positive from negative trials on non-recent compared to recent trials.Despite their greater sensitivity, YA showed the same tendency to respond negatively to all trials (response bias: c) as OA, suggesting that OA did not respond more conservatively than YA.All participants, however, tended to respond more conservatively on recent compared to non-recent trials, which may reflect in part a hesitancy to respond positively to RN trials due to the PI provoked by these trials.
Regarding brain activity, we expected to observe PI effects at frontocentral electrodes in a negativity occurring between 250 and 550 ms and in a late positivity between 400 and 960 ms after Probe stimulus onset (Tays et al., 2008;Yi & Friedman, 2014;Zhang et al., 2010), with greater positivity in non-RN compared to RN trials.Also, we expected these effects to be altered in OA.Indeed, YA and OA were found to show different patterns of brain activity upon the presentation of negative Probes in the Recent Probes task, which warranted exploring the EEG signals separately for each group and suggests that age differences in PI resolution in the Recent Probes task may be associated with the reliance on different brain mechanisms and/or cognitive strategies.

Table 3
The temporal and spatial characteristics of the temporospatial factors that showed PI effects in young (YA) and old adults (OA), in relation to their corresponding ERP components.

Targeted analysis on PI-related ERP components
Focusing on the TSFs sharing temporal and spatial characteristics with ERP components previously observed in EEG studies focused on PI effects and in their age-related differences, YA's and OA's brain activity showed TSFs that resembled either a control-related anterior N2 or MFN (Calvo & Bialystok, 2021;Llorens et al., 2020;Perfetti et al., 2014;Tays et al., 2008) or an LPC (Calvo & Bialystok, 2021;Yi & Friedman, 2014;Zhang et al., 2010).PI modulated a TSF that may represent an MFN evoked by the presentation of Probe stimuli in YA.In contrast, OA showed TSFs that may contribute to an MFN that were not modulated by PI condition.However, OA showed a central LPC that was found to be sensitive to PI condition, while in YA those TSFs that resembled the LPC did not differ between PI conditions.Tays et al. (2008) observed an MFN in their YA sample but failed to do so in the OA group, which aligns with the current results.In contrast, Tays et al. (2008) found the MFN in YA to show greater negativity in high compared to low PI trials, while the opposite holds true for the MFN observed in our YA group.Furthermore, they estimated the MFN to be generated in the left inferior frontal cortex, while in the present study the maximally activated source of the MFN was estimated to be in the right superior frontal gyrus.Tays et al. (2008) suggested that the MFN observed in their study, much like the N450 component previously reported for Stroop tasks (e.g.West et al., 2004), may have reflected the recruitment of cognitive control to overcome interference during cognitive processes.This notion is further supported by Spronk and Jonkman (2012), who found adults between 18 49 years of age to show a similar negativity with greater amplitude during a Stroop task involving face stimuli under conditions of high cognitive load induced by a concurrent n-back task.Tays et al. (2008) also found the MFN to be sensitive to the response conflict generated by stimuli in their task, suggesting that it may represent brain activity aimed at resolving cognitive conflict as well as response conflict.In the present study, MFN was estimated to be localised to the right superior frontal gyrus, which has also been previously linked with resolving cognitive conflicts as opposed to response conflicts (van Veen & Carter, 2005), and its activity has been observed to correlate with the stop signal RT (Hu et al., 2016).Hence, the MFN observed in the current study may be associated with overcoming the cognitive conflict produced by high PI Probes.
If indeed YA's MFN does represent activity similar to the MFN observed in previous research, this would suggest it also represents brain activity related with the employment of cognitive control in an attempt to overcome PI, which in the present study is disrupted when facing high PI.The MFN showed the opposite PI effect to the MFN observed by Tays et al. (2008): greater negativity in low than high PI trials.This may be explained by methodological differences between our and Tays and colleagues' study.Specifically, their task had a relatively high percentage of PI-inducing trials, which may have increased the likelihood that participants were aware of the PI manipulation.Note that this is also the case for Stroop and Stop signal tasks, as the instructions warn about the necessity to inhibit specific responses.This may have driven YA to anticipate interference and deploy some monitoring processes on all trials, which are triggered in the face of the interference or the to-be-inhibited event.This was not the case in the present study, meaning that interference detection may have triggered different, more reactive processes reflected in the lower rather than higher amplitude of the MFN.Nonetheless, this is an overly speculative explanation, and planned manipulations of interference/inhibition should be addressed

Table 4
Mean voltages (in µV) and standard deviations for the temporo-spatial factors showing between PI conditions differences for YA and OA, and the location and coordinates of the maximally activated This table presents trimmed means and non-winsorized standard deviations.Source:(a) Source for that temporo-spatial factors.(b) Source coordinates are MNI coordinates.
in future studies to shed light on these conflicting results.The present study found TSFs in both YA and OA indicative of LPCs, although modulation of an LPC was only observed in OA.In contrast, previous studies such as Zhang and colleagues (2010) and Yi and Friedman (2014) have also observed an LPC to be associated with PI in both YA and OA, albeit occurring later in OA.These authors suggested that the LPC reflects the competition between familiarity and recollection of contextual information during PI.Interestingly, as in the studies by Zhang et al. (2010) and Yi and Friedman (2014), the LPC observed in OA showed greater positivity in low compared to high PI trials.
In the present work, the LPC's maximally activated source was estimated at the right middle frontal gyrus, which is a region that has been previously found to be implicated in PI tasks.Specifically, the middle frontal gyrus has been found to show activations during modified Sternberg PI tasks that have negative, highly familiar Probes (e.g., Zhang et al., 2003).Similarly, the middle frontal gyrus has also been found to be activated in tasks that induce response conflict, such as the Flanker paradigm (Hoffman et al., 2021;Peschke et al., 2016) as well as during judgements of how recently items were memorised in an episodic memory task (Zorrilla et al., 1996).Additionally, a meta-analysis by Gavazzi et al. (2023), found the right middle frontal gyrus to be one of the key right PFC nodes involved in varying levels of reactive inhibitory control.
It has been previously suggested that PI involves a competition between recollection processes and familiarity: the familiarity of high PI trials draws participants towards making a positive response, which can be overcome by correctly recollecting the context in which it was previously seen (Oberauer, 2005).Therefore, it may be the case that the LPC reflects an executive process that signals the detection of conflict between the high familiarity and context retrieval elicited by RN Probes, despite this process not efficiently influencing behaviour in OA.However, further evidence is needed to confirm these conclusions.

Exploratory analyses of other TSF of EEG activity related to PI
While the exploratory analyses in OA did not reveal any significant differences in brain activity between PI conditions, they showed YA's TSF3.4 to be modulated by PI Condition.TSF3.4 peaked at 593.75 ms after Probe stimulus onset and had its maximally activated source estimated to be at the left medial frontal gyrus.A study by Kamiński and colleagues (2017) found neurons in the medial frontal cortex to produce sustained activity that may support WM processing.Furthermore, Kamiński and colleagues found the activity in this brain area to be modulated by WM load, and the variability in its activity over trials during WM processing was related to the accuracy and speed with which information was successfully recalled.Therefore, it may be the case that TSF3.4 represents activity recruited by YA to support WM processing during conditions requiring varying degrees of cognitive control.

Limitations
This study is not without its limitations, which should be addressed in future studies.As already mentioned, the limited time to respond to Probe stimuli (2000 ms) may have led to ceiling effects in OA's response times.Similarly, the presence of distracting stimuli during the presentation of the Target Set (i.e., the presence of faces when participants must memorise scenes, and vice versa) may have interacted with the observed effects.It may be the case that the known age-related differences in selective attention contributed to the slower responses of OA across PI conditions, masking subtle differences between PI conditions in RT data, an aspect that warrants future studies.Also, potential influences of information maintenance processes on Probe processing activity should be explored, as there is indirect evidence of potential associations (see Rominger et al., 2019).Additionally, while these results may shed light on the temporal and spatial characteristics of electrophysiological activity when overcoming PI, the limitations of source reconstruction techniques applied to scalp-recorded EEG data must be noted.Thus, the reported maximally activated sources correspond to the smoothest possible solutions, but imaging studies with tools with better spatial resolution are needed to support the present findings.Lastly, to draw firmer conclusions, the present results must be replicated using larger sample sizes.In particular, the exploratory correlational analyses require a larger sample to provide more stable estimates of the relation between PI effects in brain activity and those in behaviour.

Conclusions
In conclusion, the Recent Probes task used in the present study found response precision to be similar between groups, albeit OA showed slower responses.In contrast with previous findings, RT PI costs in the Recent Probes task were larger in YA than OA.Regarding brain activity, the present study found EEG activity to be characterised by different components in YA and OA during a Recent Probes task with varying levels of PI.This is somewhat different to previous research that found evidence to suggest that both YA and OA showed modulation of the same LPC, although this process was delayed in OA.In contrast, the current results point towards an alternative account, which suggests OA show brain activity with distinct temporal and spatial characteristics to that shown by YA in response to PI. YA showed one TSF with a negative scalp distribution over frontal and central sensors to be modulated by PI condition.We argue that this TSF represents an MFN, which has been observed in previous research.The MFN observed in the present study may represent the recruitment of cognitive control to overcome PI.Another TSF (TSF3.4) was found to differ between conditions of high and low PI in YA, potentially indicating neural activity that also supports WM processing during conditions requiring varying degrees of cognitive control.In contrast to YA, an LPC was observed in OA that was modulated by PI condition.This LPC may represent an attempt to signal the detection of conflict between familiarity and context during PI to influence behaviour.Hence, the present results suggest that OA may rely on late processes related to context recollection, while YA may rely on earlier processes related to inhibition of interference to overcome PI in Recent Probe tasks.

Sources of financial support
This study was conducted at Psychology Research Centre (UID/PSI/ 01662/2013), University of Minho, and supported by the Portuguese

Fig. 1 .
Fig. 1.Trial scheme of the Recent Probes task showing two trials from the face-relevant block.The right side of the plot indicates the duration of each trial phase.The bottom part of the figure shows what condition of PI (RN, non-RN, RP, or non-RP) would be created if any of these stimuli were presented as the Probe stimulus of the second trial.Different possible conditions of PI are achieved when the Probe matches or does not match a stimulus in the Target Set of the current and/or previous trials.Note that stimuli are images of Public Domain used exclusively for illustrative purposes.Scene images were downloaded from pdpics.com(https://pd pics.com/) and face portraits from Smithsonian Open Access (https://www.si.edu/openaccess).RN = Recent Negative.RP = Recent Positive.

Fig. 2 .
Fig. 2. Accuracy (panel A) and RT (panel B) for YA and OA in each PI Condition (non-RN and RN) across all blocks of trials.Panel C shows the proportional PI costs in RT and panel D in accuracy rates for YA and OA.Panel E shows sensitivity (d') and panel F response bias (c) for YA and OA in each PI Condition across all blocks of trials.Bars indicate non-winsorized standard errors.* * p < .01,* ** p < .001.

Fig. 3 .
Fig. 3. TSFs selected for the targeted analysis of the control-related N2-like/MFN components in YA and the original ERP data.The top row of the figure shows the original ERP topographies (left) in conditions of low (NonRN) and high (RN) PI and their difference (NonRN-RN) in the time window between 250 and 550 ms after Probe onset, as well as the Probe-locked ERP waveform averaged across Fpz, AFz and Fz frontal midline sensors (right).All other rows show the TSFs topographies (left) showing the reconstructed voltages in NonRN and RN trials and the difference between them; the sLORETA estimated source localisations of each TSF across PI conditions (central columns); and the TSF's reconstructed voltages waveform at the electrode showing peak activity (right column).TSF = temporospatial factor; YA = young adults; NonRN = non-recent negative; RN = recent negative; MFN = medial frontal negativity; * Indicates significant differences.

Fig. 4 .
Fig. 4. TSFs selected for the targeted analysis of the LPC component in YA and the original ERP data.The top row of figures shows the original ERP topographies (left) in conditions of low (NonRN) and high (RN) PI and their difference in the time window between 400 and 960 ms after Probe onset, as well as the Probe-locked ERP waveform averaged across FCz, Cz and CPz midline sensors (middle column).All other rows show the TSFs topographies (left columns) showing the reconstructed voltages in NonRN and RN trials and the difference between them; the sLORETA estimated source localisations of each TSF across PI conditions (central columns); and the TSF's reconstructed voltages waveform at the electrode showing peak activity (right column).TSF = temporospatial factor; YA = young adults; NonRN = non-recent negative; RN = recent negative; MFN = medial frontal negativity; * Indicates significant differences.

Fig. 5 .
Fig. 5. TSFs selected for the targeted analysis of the control-related N2-like/MFN components in OA and the original ERP data.The top row of the figure shows the original ERP topographies (left) in conditions of low (NonRN) and high (RN) PI and their difference (NonRN-RN) in the time window between 250 and 550 ms after Probe onset, as well as the Probe-locked ERP waveform averaged across Fpz, AFz and Fz frontal midline sensors (right).All other rows show the TSFs topographies (left) showing the reconstructed voltages in NonRN and RN trials and the difference between them; the sLORETA estimated source localisations of each TSF across PI conditions (central columns); and the TSF's reconstructed voltages waveform at the electrode showing peak activity (right column).TSF = temporospatial factor; OA = old adults; NonRN = non-recent negative; RN = recent negative; MFN = medial frontal negativity; * Indicates significant differences.

Fig. 6 .
Fig. 6.TSFs selected for the targeted analysis of the LPC component in OA and the original ERP data.The top row of figures shows the original ERP topographies (left) in conditions of low (NonRN) and high (RN) PI and their difference in the time window between 400 and 960 ms after Probe onset, as well as the Probe-locked ERP waveform averaged across FCz, Cz and CPz midline sensors (middle column).All other rows show the TSFs topographies (left columns) showing the reconstructed voltages in NonRN and RN trials and the difference between them; the sLORETA estimated source localisations of each TSF across PI conditions (central columns); and the TSF's reconstructed voltages waveform at the electrode showing peak activity (right column).TSF = temporospatial factor; OA = old adults; NonRN = non-recent negative; RN = recent negative; MFN = medial frontal negativity; * Indicates significant differences.

Table 1
YA and OA groups characterization, and t-test result comparing age, education level and MoCa scores.The values in this table are non-trimmed means and nonwinsorized standard devations.
Note.MoCA: Montreal Cognitive Assessment.Handedness was assessed by the Edinburgh Handedness Inventory

Table 2
Mean number of trials per PI Condition and Age Group and the results of an independent sample t-test comparing the number of trials in each group.The values in this table are non-trimmed means and non-winsorized standard devations.